Aims and objectives
Machine learning techniques are applied to various fields such as general object recognition. Especially in the medical imaging field,
they are mainly used for computer-aided diagnosis (CAD) [1]. In recent years deep convolutional neural network (DCNN) which is a type of machine learning has been widely used for CAD and it demonstrates high performance [2-4]. The DCNN can be used not only for general CAD that detects lesions but also for applications such as filters that remove artifacts. In addition,
it was used to remove...
Methods and materials
Figure 1 shows outline of the DLR technique. Filtering processes which remove image noise and artifacts to improve signal reliability are placed to image reconstruction processes. The DCNNs were trained by pairs of dataset,
one of the pairs was the teaching data acquired under ideal condition,
and the other was the training data including noise and artifacts. The teaching dataset were acquired with high tube current,
large number of views,
small cone beam angle,
and environment with less beam hardening effect.
We scanned 200 mm...
Results
SD value at the reference dose (100 mA) image reconstructed with hybrid IR and that at 20 mA images reconstructed with the DLR were equivalent (13.7 HU v.s.
13.2 HU) (Figure 2). This result indicated that the DLR could reduce 80% of radiation exposure from the viewpoint of SD value. However,
even if SD is equivalent,
the diagnostic capability of the image was degraded when the ratio of low frequency noise is large [6]. Figs 3 and 4 show a comparison of NPS between the...
Conclusion
From the evaluation of SD,
20% radiation dose image reconstructed by the DLR was equivalent to the reference dose image reconstructed by hybrid IR. When evaluated by noise characteristics using NPS,
even if the radiation dose of 30% was reduced,
the image reconstructed by the DLR was superior to the reference dose image reconstructed by hybrid IR. Together these results,
we concluded that the DLR could reduce at least 30% of radiation dose compared with conventional hybrid IR.
Personal information
Toru Higaki,
Ph.D.
Graduate School of Biomedical & Health Science,
Hiroshima University.
1-2-3 Kasumi,
Minami-ku,
Hiroshima,
732-8551,
Japan.
+81-82-257-5257
[email protected]
References
Erickson BJ,
Korfiatis P,
Akkus Z,
Kline TL.
Machine Learning for Medical Imaging.
RadioGraphics.
2017;37(2):505-515.
Shen D,
Wu G,
Suk HI.
Deep Learning in Medical Image Analysis.
Annu Rev Biomed Eng.
2017;19:221-248.
Chartrand G,
Cheng PM,
Vorontsov E,
et al.
Deep Learning: A Primer for Radiologists.
RadioGraphics.
2017;37(7):2113-2131.
Suzuki K.
A supervised 'lesion-enhancement' filter by use of a massive-training artificial neural network (MTANN) in computer-aided diagnosis (CAD).
Phys Med Biol.
2009;54(18):S31-45.
Chen S,
Suzuki K.
Computerized detection of lung nodules by means of "virtual dual-energy"...